Luma events

Luma Events

Overview

Use this skill for Lu.ma event work: memory-informed discovery, public event search, event page inspection, RSVP/ticket checks, calendar/community pages, host/admin workflows, guest lists, event creation, and official Luma API integrations.

Do not use this for Luma AI / Dream Machine image or video generation.

Public Boundary

This skill is safe to publish when it stays generic. Keep personal state outside the repo.

Public-safe contents:

• SKILL.md instructions.
• references/luma-api.md with official API notes and secret-handling rules.
• templates/luma-interest-profile.example.json with fake values only.
• Scripts or examples that accept a profile path but do not embed a real profile.

Never publish:

• Real memory-derived interest profiles.
• Raw memory search output, music/listening history, calendar availability, home details, tickets, QR codes, guest lists, cookies, API keys, or account screenshots.
• Account-specific defaults that imply registration, ticket ownership, exact private availability, or private travel plans.

Default Workflow

For open-ended discovery, start from a compact interest profile, then search Lu.ma from that. Public browser discovery must work even when no API key exists.

1. Build or refresh an interest profile:
   • Use the current user request first.
   • Add stable interests, location preferences, time windows, and downrank rules from approved local memory only when available.
   • Store real profiles in private state outside the public repo.
   • Use templates/luma-interest-profile.example.json only as the shape.
2. Query public Lu.ma surfaces:
   • https://luma.com/discover
   • city pages such as /seoul?k=p, /san-francisco?k=p, or /tokyo?k=p
   • category pages such as /tech?k=t, /ai?k=t, /wellness?k=t, /arts?k=t
   • relevant calendar/community slugs from search.
3. Inspect promising event pages. Do not rank from card text alone when the event is important.
4. Score candidates by:
   • direct interest match
   • location/time feasibility
   • host/community quality
   • novelty since last scan
   • actionability: RSVP open, waitlist, sold out, free/paid, online/offline.
5. Report only the top matches and notable skips unless the user asks for a full dump.

Public Discovery

For explicit search requests:

1. Parse city, date window, category, host/community, price/free preference, in-person/online preference, and purpose.
2. If under-specified but answerable, use profile defaults. Ask one focused question only when the answer would otherwise be misleading.
3. Search Lu.ma public pages and public web results.
4. Inspect strong candidate pages and extract:
   • title and canonical URL
   • host/calendar/community
   • start/end time and timezone
   • venue/city or online state
   • price/ticket/RSVP state
   • why it fits or does not fit.
5. Convert times into the event local timezone and Asia/Seoul when useful.

Event Page Inspection

For a single Lu.ma URL:

1. Fetch the page. If public extraction is thin or JS-rendered, use browser inspection.
2. Identify event identity: title, slug/event ID if visible, host, calendar/community, date/time, location, and RSVP/ticket state.
3. If the user asks whether it is "the one", separate direct proof from circumstantial matches.
4. Return a compact readout with uncertainty boundaries.

Evidence tiers:

• Direct proof: canonical Lu.ma URL/slug, official event page, authenticated dashboard/API readback, event ID, guest ID, confirmation screen, or official confirmation email.
• Circumstantial match: similar title, host, city, date, calendar name, search result, social post, profile similarity, or inferred taste fit.

RSVP / Ticket Workflow

Use browser when the user wants to RSVP, register, check a pass, or handle a waitlist.

1. Open the exact event URL in the logged-in browser profile.
2. Read visible event, ticket, and account state.
3. Before any write action, summarize the exact action and ask for confirmation.
4. For paid tickets, show price, currency, fee/refund hints if visible, event identity, and payment step before confirming.
5. After the action, verify with the Lu.ma confirmation state, ticket/pass screen, email, or calendar update when available.

Actions requiring explicit confirmation:

• RSVP/register/join waitlist
• buy tickets
• send invites
• add/remove guests
• import contacts
• create/update/cancel events
• create/update coupons
• update hosts
• approve/reject submissions
• create/update webhooks
• change membership status.

Double-confirm money-affecting or irreversible actions: event cancellation, refunds, paid purchases, paid membership approval, ownership/calendar transfer, or bulk imports.

API Workflow

Use the official API only when an API key is available and the task needs it.

1. Read references/luma-api.md.
2. Confirm whether the operation is read-only or write.
3. Keep secrets in environment variables such as LUMA_API_KEY; never echo them.
4. Test auth with GET /v1/users/get-self.
5. Fetch exact endpoint docs or the OpenAPI schema before unfamiliar writes.
6. For writes, draft the JSON payload first and ask for confirmation.
7. Read back the changed entity and compare expected fields.

Output Shape

For discovery lists:

• event name
• time and timezone
• location
• host/calendar
• matched interest vector
• price/RSVP state if known
• evidence tier
• why it fits / why it may not
• canonical URL

For actions:

• what changed
• direct proof/readback
• what was not done
• remaining user decision, if any